Skip to content

[AAAI'22] Feedback Gradient Descent: Efficient and Stable Optimization with Orthogonality for DNNs

License

Notifications You must be signed in to change notification settings

bokveizen/Feedback-Gradient-Descent

Repository files navigation

Feedback-Gradient-Descent

Bu, Fanchen and Dong Eui Chang. “Feedback Gradient Descent: Efficient and Stable Optimization with Orthogonality for DNNs.” AAAI (2022).

@inproceedings{Bu2022FeedbackGD,
  title={Feedback Gradient Descent: Efficient and Stable Optimization with Orthogonality for DNNs},
  author={Fanchen Bu and Dong Eui Chang},
  booktitle={AAAI},
  year={2022}
}

For CIFAR-10:

python main.py --model resnet --depth 28 --width 10 --optim_method FGD --lr 0.05 -lrg 0.2 --feedback 0.4 --stiefel 1 --dataset CIFAR10 --gpu_id [gpu_id] --save [log_path]

For CIFAR-100:

python main.py --model resnet --depth 28 --width 10 --optim_method FGD --lr 0.08 -lrg 0.16 --feedback 0.4 --stiefel 1 --dataset CIFAR100 --gpu_id [gpu_id] --save [log_path]

About

[AAAI'22] Feedback Gradient Descent: Efficient and Stable Optimization with Orthogonality for DNNs

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages